// Data · AI · Full-Stack
Building with data, end to end.
MS in Business Analytics & AI at UT Dallas. I design pipelines, build ML models, and ship production apps — from 700M-row/day data infrastructure to LLM agents and full-stack finance tools.
// 00. about
I'm a data and AI practitioner originally from India, now pursuing my MS in Business Analytics & AI at the University of Texas at Dallas. My background spans the full data stack — from building production pipelines and BI systems to training ML models and shipping full-stack applications.
At Cognizant, I spent 2.5 years working across data engineering and analytics — architecting PySpark pipelines processing 700M+ rows daily, designing Power BI dashboards used by leadership, and building ETL systems that eliminated manual work at scale.
I like building things that are actually used. Currently working on FinTrack, a self-hosted personal finance app with AI-powered rebalancing advice, a Telegram bot, and live portfolio tracking. I'm looking for roles where I can keep building things that matter — across data engineering, ML, or AI development.
// 01. experience
2.5 years at Cognizant spanning data engineering, analytics, and pipeline delivery at enterprise scale.
// 02. projects
Full-stack apps, ML models, LLM agents, and data visualizations.
Self-hosted, multi-user personal finance app built with Flask + Supabase. Tracks expenses, income, investments, debts, and goals — all in one place. Features a spending forecast engine (weighted 6-month prediction), financial health score across 5 pillars, stock research tab with AI analysis, broker sync (Zerodha + Groww), live price refresh (Finnhub → Yahoo → CoinGecko), and a Telegram bot powered by Groq LLaMA 3.3 70B.
Highlights: per-asset XIRR, AI spending insights with smart cache invalidation, debt tracker with 3 loan structures (EMI / Flexi / Overdraft), What-If calculator for goals, and a monthly scorecard graded A+ to F.
Multi-user health and habits tracker built with FastAPI + Supabase + SQLAlchemy. AI-powered metric extraction from smart scale photos via Groq Vision — photograph your Fitdays report and all 27 body composition metrics are extracted automatically. Features daily habit streaks, a 15-achievement system, monthly calendar heatmaps, 6 body composition trend charts, and goal tracking with live progress.
Built invite-only with middleware-level account management. Switched from Flask to FastAPI to explore async routing, Pydantic validation, and proper ORM migrations — each feature lives in its own router.
Academic advising assistant built on Google Cloud Vertex AI + Claude Agent. Uses a hybrid RAG pipeline combining Pinecone vector DB (semantic search) with Neo4j graph DB (relationship traversal) to ground LLM responses in structured academic data — reducing hallucination on course prerequisites and degree requirements.
Custom CNN in PyTorch for multi-class MRI tumor classification. Achieved 94%+ F1-score via ablation studies across optimizers (Adam vs SGD) and batch sizes. Deployed batch inference on AWS EC2 with predictions streamed into Databricks for real-time clinical review.
CNN classifying handwritten letters and digits across MNIST and USPS datasets — 98.5% accuracy. Built with PyTorch using batch normalization and dropout regularization for robust generalization.
Ensemble classification models predicting loan default risk on 50K+ applications — 98.3% accuracy. Feature engineering in Spark SQL (credit ratios, risk tiers) with class-imbalance handling. Insights delivered via Power BI backed by Delta Lake.
Predicted water potability using six ML algorithms — accuracy ranging from 92.2% to 99.8% across models. Comprehensive EDA and visualization on Kaggle dataset; hyperparameter tuning via RandomizedSearchCV.
Predictive model forecasting Kickstarter campaign success. Compared Decision Tree, Random Forest, Gradient Boost, and AdaBoost — achieving up to 85.78% accuracy with hyperparameter tuning.
Supervised ML model and web application detecting plant NPK deficiencies from images — 87.6% accuracy. Built end-to-end from model training through deployment as an interactive web app.
End-to-end retail analytics — market basket analysis (MLxtend), customer segmentation via K-means clustering, and regression modeling on transaction data to improve targeted marketing.
Tableau dashboard comparing 15-year returns of Mutual Funds vs Market Index. Formulated an optimal asset allocation strategy to achieve a 20% CAGR target across index instruments.
Power BI dashboard providing executive-level visibility into sales, profit, orders, and profit margin — with drill-through capability across product lines and time periods.
// 03. skills
Spanning data engineering, analytics, ML, full-stack, and cloud infrastructure.
// 04. certifications
// 05. education
// 06. blog
Notes on data engineering, ML, and things I've learned building in the field.
// 07. contact
Open to data engineering, ML, and AI opportunities. Let's talk.